ZUG, Switzerland, October 17, 2023 (Newswire.com) - LALAL.AI, a cross-platform AI-powered stem splitting tool, presents Orion, the algorithm that delivers a leap forward in audio separation technology, boasting capabilities in extracting vocals from a mix in a much cleaner and faster way than all existing solutions. 

Unlike other stem splitting AI tools, Orion hallucinates the vocals and directly synthesizes them, not relying on a spectral mask that can’t deliver anything more than in the original track, which is a breakthrough approach to audio separation. It means the new algorithm doesn't just extract vocals from a source, it assumes what a vocal stem is before mixing, and, if it’s not entirely suitable for extraction, recreates the vocal stem. 

The peculiarities of the Orion algorithm are as follows: 

• Orion significantly reduces the phasing effect and distortion in the extracted stems compared to the older generations of the LALAL.AI algorithms. 
• The results split by Orion are 70% cleaner from various artifacts.
• Orion preserves all tonal characteristics of a vocal stem. 
• Orion can accurately extract unison singing and echoes. 

“We spent 20 times more training data on Orion than we did on Phoenix, our previous solution. Besides, in addition to all the differences from our former algorithms quality-wise, Orion is twice as fast as Phoenix, meaning that users will get the split results much quicker, especially in batch processing,” says Lead Engineer Nikolay Pogorskiy. 

To evaluate the separation quality, the LALAL.AI engineers use signal-to-distortion ratio (SDR), indicating how many artifacts the separation process introduces to the vocal and instrumental parts. The difference between Orion and Phoenix is 2.5 dB, which is a huge leap in quality in comparison to the SDR differences between the previous algorithms. 

Sound producers, DJs, musicians, and other creatives can now test Orion, as well as compare it with the previous algorithm on the LALAL.AI website. The new solution can currently be used for the extraction of vocals, instrumental backtracks, and noise cancellation. More instruments, such as acoustic and electric guitars, drums, bass, piano, synthesizer, strings, and winds can be processed by Orion in the upcoming releases. 

To learn more about Orion and listen to the samples split by it, check the LALAL.AI blog article. 

Contact Information:
Klara Alexeeva
Marketing Manager
[email protected]


Original Source: LALAL.AI Releases Orion, AI That Provides Studio-Level Vocal Extraction Quality